Neutron-antineutron oscillation sim/reco needs Georgia Karagiorgi, - - PowerPoint PPT Presentation

Neutron-antineutron oscillation sim/reco needs

Georgia Karagiorgi, Yuyang Zhou, Jeremy Hewes Joint NDK/High-E and FD Sim/Reco DUNE Physics Week

• Nov. 2017

Neutron-antineutron oscillation in DUNE

Search for rare, baryon-number violating signature. Anticipated background contributions from atmospheric neutrinos. Signature/topology is visually striking: “star event” à Use a trained CNN to differentiate n- nbar events from atmospheric neutrino events. For more details on analysis method, results, and current status, see yesterday’s talk by Yuyang Zhou:

https://indico.fnal.gov/event/15181/session/2/contribution/11/material/ slides/0.pdf

Neutron-antineutron oscillation in DUNE

Results: Network training results show excellent signal vs. background separation. Optimized score cut yields 14% signal efficiency and 99.997% background rejection efficiency à DUNE sensitivity is 5x better than the existing bound from Super-Kamiokande (leading world limit on this process). Systematic uncertainties: largely educated guesses based on Super-K search e.g. CNN score cut signal selection background rejection

Simulation needs

Systematic uncertainty assumptions are in need of further work and studies to validate/ better quantify. Signal simulation uncertainties:

• nuclear suppression factor for bound-neutron oscillation
• final state branching fractions and final state interactions within the nucleus

Background simulation uncertainties:

• atmospheric neutrino fluxes
• neutrino cross-sections, including nuclear effects and final state interactions within

the nucleus Selection efficiency:

• ROI selection efficiency
• finite statistics size of sample used for training and inference
• backgrounds from other detector activity (e.g. cosmics)

Detector response:

• gain variation
• field non-uniformity
• electron lifetime
• noise levels

Simulation needs

Systematic uncertainty assumptions are in need of further work and studies to validate/ better quantify. Signal simulation uncertainties:

• nuclear suppression factor for bound-neutron oscillation
• final state branching fractions and final state interactions within the nucleus

Background simulation uncertainties:

• atmospheric neutrino fluxes
• neutrino cross-sections, including nuclear effects and final state interactions within

the nucleus

• backgrounds from other detector activity (e.g. cosmics)

Selection efficiency:

• APA and ROI selection
• finite statistics of samples used for training and inference

Detector response:

• gain variation
• field non-uniformity
• electron lifetime
• noise levels
• data reduction scheme

from theory; can quote bound lifetime

• J. Barrow et al.

~ well-characterized ~ well-characterized need dedicated studies à more event samples!

Reconstruction needs

• Very minimal reconstruction needs (we use recob::Wire after deconvolution

and before hit finding), and then apply APA/ROI selection and feed images to CNN.

• However, it might be useful to study different zero suppression/data reduction

schemes à deconvolution; and effects of different ROI selection (threshold based, after deconvolution). For each MC systematic variation, will need:

• ~200,000 n-nbar events and ~200,000 atmospheric neutrino events (for

inference)

• 4,000 jobs x 5,000 events/job, on grid, taking about 5,000 hrs per sample, and

200-500 GB per sample For additional backgrounds studies, will need

• Cosmogenic simulations (10x expected data statistics?)

processing